Charging pile power load prediction and evaluation system and evaluation method

Abstract

The invention relates to the technical field of power load prediction, in particular to a charging pile power load prediction and evaluation system and evaluation method. Comprising a data acquisition module, a model prediction module, a dynamic adjustment module and a fault alarm module. Historical charging power, charging duration and charging frequency data of a charging pile are acquired through a data acquisition module, a model prediction module establishes a power load prediction model of the charging pile by combining a convolutional neural network with a long-short-term memory network, and a dynamic adjustment module uses a genetic optimization algorithm to adjust the power load of the charging pile. The optimal power distribution scheme of each charging pile in different time periods is calculated, the operation states of the charging piles are dynamically adjusted through the intelligent charging management system, and the fault alarm module establishes a graded early warning system, so that the stability of power supply is ensured, early warning information is sent in advance, workers are notified to perform maintenance, and potential safety hazards are eliminated.

Description

Technical Field

[0001] The present invention relates to the technical field of electric load forecasting, and more specifically, to a charging pile electric load forecasting and evaluation system and an evaluation method. Background Art

[0002] With the popularization of electric vehicles, the number of charging piles has increased rapidly. The power of different types of charging piles varies greatly. For example, the power of common AC slow charging piles is generally about 7kW, while the power of DC fast charging piles can reach 100kW or even higher. Accurately forecasting the electric load of charging piles is crucial for the stable operation of the power system and the rational allocation of resources. The charging pile electric load forecasting and evaluation system has emerged. It uses the technical means of reinforcement learning to monitor, analyze and forecast the power usage of charging piles in real time, providing decision-making support for power departments and related operating enterprises.

[0003] Currently, with the rapid increase in the number of charging piles, when a large number of high-power charging piles work simultaneously, it will instantaneously increase the power demand of the local power grid, leading to a tight power distribution in this area and affecting the normal power supply of other electrical equipment. At the same time, the charging time of users often concentrates in periods such as after work and at night, and it will overlap with residential and commercial electricity consumption during peak electricity consumption periods, further increasing the peak-valley difference of the power load and resulting in insufficient power supply. In order to be able to forecast the electric load of charging piles based on the charging power, charging duration and charging times of charging piles, dynamically adjust the power distribution according to the forecast results, timely detect abnormal load changes and potential power failure risks, and send early warning information in advance. Therefore, we propose a charging pile electric load forecasting and evaluation system and an evaluation method. Summary of the Invention

[0004] The purpose of the present invention is to solve the problem that when a large number of high-power charging piles work simultaneously, it will instantaneously increase the power demand of the local power grid, leading to a tight power distribution in this area and affecting the normal power supply of other electrical equipment. In order to be able to forecast the electric load of charging piles based on the charging power, charging duration and charging times of charging piles, dynamically adjust the power distribution according to the forecast results, timely detect abnormal load changes and potential power failure risks, and send early warning information in advance.

[0005] To achieve the above object, the present invention provides a charging pile electric load forecasting and evaluation system, including a data acquisition module, a model prediction module, a dynamic adjustment module and a fault alarm module;

[0006] The data acquisition module collects the historical charging power, charging duration and charging times data of the charging pile by accessing the data recording and storage module of the charging pile, arranges them in chronological order using the timestamp information in the data to form time series data, and transfers the time series data to the model prediction module;

[0007] The model prediction module divides the time series data transmitted by the data acquisition module into the same time steps, extracts local features from the input data using a convolutional neural network, and then inputs the extracted features into a long short-term memory network for time series modeling to establish a power load prediction model for the charging pile. The Adam optimization algorithm is used to evaluate the predicted power load value of the charging pile with the mean squared error as the metric;

[0008] The dynamic adjustment module aims to ensure the stability of power supply. According to the power load value of the charging pile predicted by the model prediction module, it uses a genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile at different time periods, and dynamically adjusts the operating state of the charging pile through an intelligent charging management system, interacts with the grid dispatching system in real time, and coordinates the power supply plan of the grid and the charging arrangement of the charging pile according to the overall load situation of the grid and the power demand of the charging pile;

[0009] The fault alarm module compares the power load value of the charging pile predicted by the model prediction module with the theoretical power load value, establishes a hierarchical early warning system according to the degree of abnormality and the size of potential risks, and takes different response measures and notification methods for different levels of early warnings.

[0010] Preferably, the model prediction module includes a model establishment unit and a training optimization unit;

[0011] The model establishment unit takes the time series data with the same step length as the input feature and the predicted power charge of the charging pile as the output result, and establishes a power load prediction model for the charging pile by combining a convolutional neural network and a long short-term memory network;

[0012] The training optimization unit trains the model with the training set data, records the changes in the loss value and evaluation metrics, draws the loss curve and evaluation metric curve, and trains and optimizes the model.

[0013] Preferably, the model establishment unit constructs a convolutional neural network layer by adding a convolutional layer at the beginning of the model and a pooling layer after the convolutional layer, takes the output of the neural network layer as the input of the long short-term memory network layer, and adds a fully connected layer after the long short-term memory network layer to determine the structure of the power charge prediction model of the charging pile.

[0014] Preferably, the model establishment unit uses the mean squared error as the loss function to measure the difference between the predicted value and the true value, and its formula is:

[0015]

[0016] where MSE is the mean squared error, n is the number of samples, y iis the true value, is the predicted value.

[0017] Preferably, the training optimization unit uses the K-fold cross-validation method to determine the number of training rounds and batch size of the charging pile power charge prediction model, and uses the grid search method to adjust the hyperparameters of the charging pile power charge prediction model.

[0018] Preferably, the dynamic adjustment module includes a strategy formulation unit and a remote adjustment unit;

[0019] The strategy formulation unit aims to ensure the stability of power supply, takes the capacity limit of the power grid, the power limit of the charging pile, and the charging demand of users as constraints, and uses the genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile at different time periods;

[0020] The remote adjustment unit communicates with the charging pile in real time through a communication protocol, dynamically adjusts the operating state of the charging pile, establishes a two-way communication channel with the power grid dispatching system, and reports the power consumption demand, charging status, and charging plan of the charging pile to the power grid dispatching system in real time.

[0021] Preferably, when the strategy formulation unit uses the genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile at different time periods, it takes minimizing the power grid load fluctuation as the objective function and determines the fitness function. The formula is:

[0022]

[0023] where F is the fitness function, T is the total number of time periods, L(t) is the total load within time period t, is the average load, and ε is a positive number.

[0024] Preferably, the remote adjustment unit uses the distributed cooperative optimization algorithm to globally optimize the charging process of the charging pile.

[0025] Preferably, the fault alarm module uses time as the horizontal axis, and the predicted power load value and the theoretical power load value as the vertical axis respectively to draw two line charts, and compares the predicted charging pile power load value with the theoretical power load value.

[0026] The second object of the present invention is to provide a method for predicting and evaluating the power load of a charging pile, including the charging pile power load prediction and evaluation system described in any one of the above, and including the following steps:

[0027] S1. The data acquisition module collects the historical charging power, charging duration, and charging times data of the charging pile, and uses the timestamp information in the data to arrange them in chronological order to form time series data;

[0028] S2. The model prediction module combines a convolutional neural network and a long short-term memory network to establish a power load prediction model for the charging pile and evaluate the predicted power load value of the charging pile;

[0029] S3. The dynamic adjustment module uses a genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile at different time periods and dynamically adjusts the operating status of the charging pile through an intelligent charging management system;

[0030] S4. The fault alarm module compares the predicted power load value of the charging pile with the theoretical power load value and establishes a hierarchical early warning system according to the degree of abnormality and the size of potential risks.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] 1. For the power load prediction and evaluation system and method of the charging pile, the data acquisition module collects the historical charging power, charging duration, and charging times data of the charging pile, and uses the timestamp information in the data to arrange them in chronological order to form time series data. The model prediction module combines a convolutional neural network and a long short-term memory network to establish a power load prediction model for the charging pile, extracts the local features of the power load data through the convolutional neural network, and uses the time series information of the long short-term memory network to capture long-term dependence relationships, realizing accurate prediction of the power charge of the charging pile;

[0033] 2. The dynamic adjustment module calculates the optimal power distribution plan for each charging pile at different time periods using a genetic optimization algorithm based on the predicted value of the power charge of the charging pile, and dynamically adjusts the operating status of the charging pile through an intelligent charging management system. According to the overall load situation of the power grid and the power demand of the charging pile, it coordinates the power supply plan of the power grid and the charging arrangement of the charging pile to ensure the normal power supply of other electrical equipment in this area and ensure the stability of power supply;

[0034] 3. The fault alarm module compares the predicted power load value of the charging pile with the theoretical power load value and establishes a hierarchical early warning system according to the degree of abnormality and the size of potential risks, timely discovers abnormal load changes and potential power fault risks, issues early warning information in advance, and notifies the staff to carry out repairs to eliminate potential safety hazards. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is the overall flow block diagram of the present invention;

[0036] Figure 2 is the overall detailed flow chart of the present invention;

[0037] Figure 3 is the method flow chart of the present invention.

[0038] The meanings of each label in the figure are as follows:

[0039] 100, data acquisition module; 200, model prediction module; 210, model establishment unit; 220, training and optimization unit; 300, dynamic adjustment module; 310, strategy formulation unit; 320, remote adjustment unit; 400, fault alarm module. Specific implementation manner

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0041] At present, when a large number of high-power charging piles work simultaneously, it will instantaneously increase the power demand of the local power grid, resulting in a tense power distribution in this area and affecting the normal power supply of other electrical equipment. In order to be able to predict the power load of the charging piles according to the charging power, charging duration and charging times of the charging piles, and dynamically adjust the power distribution according to the prediction results, timely detect abnormal load changes and potential power failure risks, and send out early warning information in advance.

[0042] Therefore, the present invention proposes to collect the historical charging power, charging duration and charging times data of the charging piles through the data acquisition module. The model prediction module combines the convolutional neural network and the long short-term memory network to establish a power load prediction model for the charging piles. The dynamic adjustment module uses the genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile at different time periods, and dynamically adjusts the operating state of the charging piles through the intelligent charging management system. The fault alarm module compares the predicted power load value of the charging pile with the theoretical power load value, and establishes a hierarchical early warning system according to the degree of abnormality and the size of the potential risk.

[0043] Specifically as follows:

[0044] As Figure 1 shown, one of the purposes of the present invention is to provide a power load prediction and evaluation system for charging piles, including a data acquisition module 100, a model prediction module 200, a dynamic adjustment module 300 and a fault alarm module 400;

[0045] The data acquisition module 100 collects the historical charging power, charging duration and charging times data of the charging piles by accessing the data record and storage module of the charging piles, arranges them in chronological order using the timestamp information in the data to form time series data, and transmits the time series data to the model prediction module 200;

[0046] The model prediction module 200 divides the time series data transmitted by the data acquisition module 100 into the same time steps, extracts local features from the input data using a convolutional neural network, and then inputs the extracted features into a long short-term memory network for time series modeling to establish a power load prediction model for the charging pile. The Adam optimization algorithm is used to evaluate the predicted power load value of the charging pile with the mean square error as the index;

[0047] The dynamic adjustment module 300 aims to ensure the stability of power supply. According to the power load value of the charging pile predicted by the model prediction module 200, it uses a genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile in different time periods, and dynamically adjusts the operating state of the charging pile through an intelligent charging management system, interacts with the power grid dispatching system in real time, and coordinates the power supply plan of the power grid and the charging arrangement of the charging pile according to the overall load situation of the power grid and the power demand of the charging pile;

[0048] The fault alarm module 400 compares the power load value of the charging pile predicted by the model prediction module 200 with the theoretical power load value, establishes a hierarchical early warning system according to the degree of abnormality and the size of potential risks, and takes different response measures and notification methods for different levels of early warnings;

[0049] Extract historical charging data from the charging pile management system, database or related monitoring devices to ensure that the data covers the three key indicators of charging power, charging duration and charging times, and try to obtain data with a longer time span to ensure that the model can learn the characteristics and laws of different time periods;

[0050] The processed data is divided into input sequences and corresponding target values according to a certain time step. For example, if the charging power of the next time step is to be predicted based on the data of the past m time steps, the data of every consecutive m time steps is used as an input sequence, and the charging power of the next time step is used as the target value.

[0051] As Figure 2 shown, among them, the model prediction module 200 includes a model establishment unit 210 and a training and optimization unit 220;

[0052] The model establishment unit 210 uses the time series data with the same time step as the input feature and the predicted power charge of the charging pile as the output result, and establishes a power load prediction model for the charging pile by combining a convolutional neural network and a long short-term memory network;

[0053] The training and optimization unit 220 trains the model with the training set data, records the changes in the loss value and evaluation index, draws the loss curve and evaluation index curve, and trains and optimizes the model;

[0054] The processed time series data is divided into a training set, a validation set, and a test set according to a certain ratio. The division ratio is 70% for the training set, 15% for the validation set, and 15% for the test set. The training set is used for parameter learning of the model, the validation set is used to adjust the hyperparameters of the model, and the test set is used to evaluate the final performance of the model.

[0055] In order to better determine the structure of the prediction model, among them, the model unit 210 is established by adding a convolutional layer at the beginning of the model, adding a pooling layer after the convolutional layer, constructing a convolutional neural network layer, using the output of the neural network layer as the input of the long short-term memory network layer, and adding a fully connected layer behind the long short-term memory network layer to determine the structure of the power charge prediction model for the charging pile;

[0056] Add a convolutional neural network layer (CNN) at the beginning of the model to extract local features of the power load data. Use a one-dimensional convolutional layer (Conv1D) to capture different features of the data through different convolutional kernel sizes and numbers. A pooling layer (such as a max pooling layer) can be added after the convolutional layer to reduce the data dimension and computational amount;

[0057] Use the output of the convolutional neural network layer (CNN) as the input of the long short-term memory network layer (LSTM). The LSTM layer further processes the time series information of the data and captures long-term dependencies. Multiple LSTM layers can be added as needed to improve the expressive ability of the model;

[0058] Add a fully connected layer after the LSTM layer to map the output of the LSTM layer to the final predicted value. The structure and activation function selection of the fully connected layer are similar to those of the LSTM model.

[0059] In order to measure the difference between the predicted value and the true value, among them, the model unit 210 uses the mean squared error as the loss function to measure the difference between the predicted value and the true value. Its formula is:

[0060]

[0061] Among them, MSE is the mean squared error, n is the number of samples, y i is the true value, is the predicted value.

[0062] The mean squared error (MSE) refers to the average of the squares of the differences between the predicted value and the true value. During the training process of models such as neural networks, first through forward propagation, the input data passes through the calculations of each hidden layer and the action of the activation function, and finally the predicted value of the output layer is obtained For each sample, the corresponding predicted value is calculated according to the structure and parameters of the model;

[0063] According to the above formula, the true value y of each sample i and the predicted value are substituted to calculate the sum of squared errors of all samples, and then divided by the number of samples n to obtain the mean squared error MSE. This value reflects the overall fitting degree of the model to the training data under the current parameters. The smaller the value, the closer the predicted value is to the true value, and the better the performance of the model.

[0064] In order to better determine the parameters of the prediction model, among them, the training optimization unit 220 uses the K-fold cross-validation method to determine the number of training rounds and batch size of the charging pile power charge prediction model, and uses the grid search method to adjust the hyperparameters of the charging pile power charge prediction model;

[0065] The K-fold cross-validation method is an effective model evaluation method that can fully evaluate the model performance on limited data, avoid overfitting, and help determine the optimal number of training rounds and batch size of the charging pile power load prediction model;

[0066] The value of K can usually be selected as 5 or 10, which are relatively common and have been verified to have good effects. The selection of the value of K needs to be weighed between the calculation cost and the evaluation accuracy. The larger the value of K, the relatively smaller the amount of data in each validation set, and the closer the evaluation result is to the true performance of the model on the entire data set, but the calculation cost will also increase accordingly. The smaller the value of K, the faster the calculation speed, but the accuracy of the evaluation result may be affected to a certain extent;

[0067] Determine a reasonable range of the number of training rounds, for example, from 10 to 100, and try at intervals of 10, that is, set as [10, 20, 30, …, 100]. If the number of training rounds is too small, the model may not be able to fully learn the data features. If the number of training rounds is too large, it may lead to overfitting;

[0068] Select a series of possible batch sizes, such as [8, 16, 32, 64]. The batch size determines the number of samples used when updating the model parameters each time. A smaller batch size can increase the randomness of model training, help jump out of the local optimal solution, but will increase the training time. A larger batch size can speed up the training speed, but may cause the model to converge to the local optimal.

[0069] Use the grid search method to adjust the hyperparameters of the prediction model. For each hyperparameter of the model, such as the convolution kernel size of the CNN and the number of units of the LSTM, determine a reasonable value range, combine the values of each hyperparameter to form different parameter combinations, evaluate each parameter combination using cross-validation, calculate the evaluation metrics on the validation set, compare the evaluation results of all parameter combinations, and select the parameter combination that makes the evaluation metrics optimal as the best parameters of the model.

[0070] Among them, the dynamic adjustment module 300 includes a strategy formulation unit 310 and a remote adjustment unit 320;

[0071] The strategy formulation unit 310 aims to ensure the stability of power supply. Taking the capacity limit of the power grid, the power limit of the charging piles, and the charging demands of users as constraint conditions, it uses the genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile at different time periods;

[0072] The remote adjustment unit 320 conducts real-time communication with the charging piles through a communication protocol, dynamically adjusts the operating states of the charging piles, establishes a two-way communication channel with the power grid dispatching system, and reports the power consumption demands, charging states, and charging plans of the charging piles to the power grid dispatching system in real time;

[0073] Through the intelligent charging management system, remotely control the output power and charging time of the charging piles. According to the power distribution plan, adjust the operating states of the charging piles in real time, realize the reasonable distribution of power, conduct real-time interaction with the power grid dispatching system, and coordinate the power supply plan of the power grid and the charging arrangements of the charging piles according to the overall load situation of the power grid and the power demands of the charging piles. When the power grid load is tight, the charging power of the charging piles can be appropriately reduced or the charging of some charging piles can be suspended to ensure the stable operation of the power grid.

[0074] In order to better determine the fitness function, among them, when the strategy formulation unit 310 uses the genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile at different time periods, it takes minimizing the power grid load fluctuation as the objective function to determine the fitness function, and its formula is:

[0075]

[0076] Among them, F is the fitness function, T is the total number of time periods, L(t) is the total load within time period t, is the average load, and ε is a positive number.

[0077] Encode the power distribution plan of each charging pile at different time periods into a chromosome, randomly generate a group of initial power distribution plans as a population, calculate the fitness value of each chromosome according to the optimization objective, select the chromosomes with high fitness through the selection operation, perform crossover and mutation operations to generate new chromosomes, and repeat the above steps until the termination condition is met (such as reaching the maximum number of iterations or the fitness value converges);

[0078] Analyze the obtained optimal power distribution plan, including the charging power changes of each charging pile at different time periods, the power grid load distribution, etc., evaluate whether the plan meets the requirements of power supply stability and various constraint conditions, and at the same time, it can be compared with other optimization algorithms or traditional distribution plans to verify the effectiveness of the genetic optimization algorithm.

[0079] In order to better dynamically regulate the behavior state of the charging piles, among which, the remote adjustment unit 320 uses a distributed collaborative optimization algorithm to globally optimize the charging process of the charging piles;

[0080] The charging piles are divided into different sub-regions or clusters, and the charging piles within each sub-region communicate with each other to exchange information such as charging status and power demand. For example, several adjacent charging piles can form a cluster, and they share data through a wireless communication network;

[0081] Each charging pile or sub-region uses an optimization algorithm to perform local optimization calculations based on the received information to determine the optimal charging power or charging strategy at the current moment. For example, in DPSO, each particle (representing a charging pile or sub-region) updates its speed and position according to its own historical optimal position and the optimal positions of neighboring particles, that is, updates the charging power;

[0082] Each sub-region or charging pile uploads the local optimization results to the superior management node or the central server, and the central server summarizes and analyzes this information to obtain the global charging status information;

[0083] Through a specific coordination mechanism, the optimization results of each sub-region or charging pile are adjusted and coordinated to make the entire charging system develop towards the global optimum direction. For example, when it is found that the grid load in a certain area is too high, the central server notifies the charging piles in that area to reduce the charging power, and at the same time adjusts the charging plans of the charging piles in other areas to balance the grid load.

[0084] In order to more intuitively compare the predicted power load value of the charging pile with the theoretical power load value, among which, the fault alarm module 400 uses time as the horizontal axis, and the predicted power load value and the theoretical power load value as the vertical axis respectively to draw two line charts to compare the predicted power load value of the charging pile with the theoretical power load value;

[0085] According to the results of data analysis, determine the key indicators for early warning, such as load growth rate, load volatility, and deviation rate from historical data, and comprehensively consider the changes of these indicators to formulate a comprehensive early warning rule. For example, when the load growth rate exceeds 20% and the load volatility exceeds 15%, an early warning message is issued;

[0086] According to the degree of abnormality and the size of potential risks, the early warning information is divided into different levels, such as general early warning, important early warning, and emergency early warning. For early warnings of different levels, different response measures and notification methods are taken. General early warnings can be notified to relevant personnel through text messages or APP push, important early warnings require a phone call notification and require relevant personnel to pay attention, and emergency early warnings require immediate measures, such as stopping the operation of some charging piles for a comprehensive inspection;

[0087] Using a data monitoring platform, the predicted power load data and the changes in various warning indicators are displayed in real time. When the indicators reach the warning threshold, the system automatically sends out warning messages and pushes the warning messages to relevant operation and maintenance personnel, management personnel, and technical experts. At the same time, the warning messages are recorded and tracked on the monitoring platform for subsequent analysis and processing.

[0088] The second object of the present invention is to provide a method for predicting and evaluating the power load of charging piles, including the charging pile power load prediction and evaluation system according to any one of the above, comprising the following steps:

[0089] S1. The data acquisition module 100 collects the historical charging power, charging duration, and charging times data of the charging piles, and arranges them in chronological order using the timestamp information in the data to form time series data;

[0090] S2. The model prediction module 200 combines a convolutional neural network and a long short-term memory network to establish a power load prediction model for the charging piles and evaluate the predicted power load values of the charging piles;

[0091] S3. The dynamic adjustment module 300 uses a genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile in different time periods, and dynamically adjusts the operating status of the charging piles through an intelligent charging management system;

[0092] S4. The fault alarm module 400 compares the predicted power load values of the charging piles with the theoretical power load values, and establishes a hierarchical warning system according to the degree of abnormality and the size of potential risks.

[0093] In summary, the working principle of this solution is as follows:

[0094] The power load prediction and evaluation system and method for the charging pile collect the historical charging power, charging duration and charging times data of the charging pile through the data acquisition module 100. Using the timestamp information in the data, it arranges them in chronological order to form time series data. The model prediction module 200 combines a convolutional neural network and a long short-term memory network to establish a power load prediction model for the charging pile. It extracts the local features of the power load data through the convolutional neural network and utilizes the time series information of the long short-term memory network to capture long-term dependencies, achieving accurate prediction of the power charge of the charging pile. The dynamic adjustment module 300 calculates the optimal power distribution plan for each charging pile at different time periods using the genetic optimization algorithm based on the predicted value of the power charge of the charging pile, and dynamically adjusts the operating state of the charging pile through the intelligent charging management system. According to the overall load situation of the power grid and the power demand of the charging pile, it coordinates the power supply plan of the power grid and the charging arrangement of the charging pile to ensure the normal power supply of other electrical equipment in this area and ensure the stability of the power supply. The fault alarm module 400 compares the predicted power load value of the charging pile with the theoretical power load value, establishes a hierarchical early warning system according to the degree of abnormality and the size of potential risks, discovers abnormal load changes and potential power fault risks in a timely manner, and sends out early warning information in advance to notify the staff for maintenance and eliminate potential safety hazards.

[0095] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. Charging pile power load prediction and evaluation system, characterized by: It comprises a data acquisition module (100), a model prediction module (200), a dynamic adjustment module (300) and a fault alarm module (400); The data acquisition module (100) collects historical charging power, charging duration and charging times data of the charging pile by accessing the data recording and storage module of the charging pile, arranges the data in chronological order using the timestamp information in the data to form time series data, and transmits the time series data to the model prediction module (200); The model prediction module (200) divides the time series data transmitted by the data acquisition module (100) into the same time steps, uses a convolutional neural network to extract local features of the input data, and then inputs the extracted features into a long short-term memory network to perform time series modeling, establish a charging pile power load prediction model, and use the Adam optimization algorithm to evaluate the charging pile power load prediction value using the mean square error as an indicator; The dynamic adjustment module (300) aims to ensure the stability of power supply, and uses a genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile in different time periods based on the power load value of the charging pile predicted by the model prediction module (200), and dynamically adjusts the operating state of the charging pile through an intelligent charging management system, interacts with the power grid dispatching system in real time, and coordinates the power supply plan of the power grid and the charging arrangement of the charging pile according to the overall load of the power grid and the power demand of the charging pile. The fault alarm module (400) compares the charging pile power load value predicted by the model prediction module (200) with the theoretical power load value, and establishes a graded warning system according to the degree of abnormality and the size of potential risks.

2. The charging pile power load prediction and evaluation system according to claim 1 is characterized in that: The model prediction module (200) includes a model building unit (210) and a training optimization unit (220); The model building unit (210) uses time series data of the same step length as input features and predicts the electric charge of the charging pile as output results, and builds a charging pile electric load prediction model by combining a convolutional neural network with a long short-term memory network; The training optimization unit (220) trains the model using training set data, records changes in loss values ​​and evaluation indicators, draws loss curves and evaluation indicator curves, and performs training optimization on the model.

3. The charging pile power load prediction and evaluation system according to claim 2 is characterized in that: The model building unit (210) adds a convolution layer to the initial part of the model, adds a pooling layer after the convolution layer, constructs a convolution neural network layer, uses the output of the neural network layer as the input of the long short-term memory network layer, and adds a fully connected layer after the long short-term memory network layer to determine the structure of the electric charge prediction model of the charging pile.

4. The charging pile power load prediction and evaluation system according to claim 2 is characterized in that: The model building unit (210) uses mean square error as a loss function to measure the difference between the predicted value and the true value, and its formula is: Among them, MSE is the mean square error, n is the number of samples, and y i is the true value, is the predicted value.

5. The charging pile power load prediction and evaluation system according to claim 2 is characterized in that: The training optimization unit (220) uses a K-fold cross validation method to determine the number of training rounds and batch size of the charging pile power charge prediction model, and uses a grid search method to adjust the hyperparameters of the charging pile power charge prediction model.

6. The charging pile power load prediction and evaluation system according to claim 1, characterized in that: The dynamic adjustment module (300) comprises a strategy formulation unit (310) and a remote adjustment unit (320); The strategy formulation unit (310) aims to ensure the stability of power supply, takes the capacity limit of the power grid, the power limit of the charging pile and the charging demand of the user as constraints, and uses a genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile in different time periods; The remote adjustment unit (320) communicates with the charging pile in real time through a communication protocol, dynamically adjusts the operating state of the charging pile, establishes a two-way communication channel with the power grid dispatching system, and reports the power demand, charging state and charging plan of the charging pile to the power grid dispatching system in real time.

7. The charging pile power load prediction and evaluation system according to claim 6 is characterized in that: The strategy formulation unit (310) uses a genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile in different time periods, taking minimizing the grid load fluctuation as the objective function and determining the fitness function, the formula of which is: Among them, F is the fitness function, T is the total number of time periods, L(t) is the total load in time period t, is the average load, and ε is a positive number.

8. The charging pile power load prediction and evaluation system according to claim 6 is characterized in that: The remote adjustment unit (320) uses a distributed collaborative optimization algorithm to globally optimize the charging process of the charging pile.

9. The charging pile power load prediction and evaluation system according to claim 1, characterized in that: The fault alarm module (400) draws two line graphs with time as the horizontal axis and the predicted power load value and the theoretical power load value as the vertical axis, respectively, to compare the predicted charging pile power load value with the theoretical power load value.

10. A method for implementing a charging pile power load prediction and evaluation method, comprising a charging pile power load prediction and evaluation system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1, a data collection module (100) collects historical charging power, charging duration and charging times data of a charging pile, and arranges the data in chronological order using timestamp information in the data to form time series data; S2, the model prediction module (200) uses a convolutional neural network combined with a long short-term memory network to establish a power load prediction model for the charging pile, and evaluates the power load prediction value of the charging pile; S3, the dynamic adjustment module (300) uses a genetic optimization algorithm to calculate the optimal power distribution plan for each charging pile in different time periods, and dynamically adjusts the operating state of the charging pile through an intelligent charging management system; S4. The fault alarm module (400) compares the predicted charging pile power load value with the theoretical power load value, and establishes a graded warning system according to the degree of abnormality and the size of potential risks.

Images